Automated sternocleidomastoid ROI placement on carotid black-blood T1-weighted MRI: toward standardized plaque-to-muscle ratio assessment.

Abstract

In carotid black-blood T1-weighted imaging (BB T1WI), plaque-to-muscle ratio (PMR) is used for quantitative assessment of plaque hyperintensity. However, sternocleidomastoid muscle (SCM) signal measurement depends on reader region of interest (ROI) placement, which may reduce reproducibility. We aimed to develop and evaluate a segmentation-based automated pipeline for SCM ROI placement and signal measurement to standardize PMR assessment. In this study, 66 patients were used for model development, patient-level 5-fold cross-validation, and final model training, and an independent historical cohort of 60 patients was used for independent ROI testing. BB T1WI was acquired on a 3.0-T scanner. A 2.5D U-Net segmented bilateral SCM on standardized axial reformats. Automated circular ROIs were placed within segmented SCM on three predefined slices centered at the carotid bifurcation. Segmentation performance was assessed using Dice, intersection-over-union, and average surface distance. Agreement between automated and manual SCM measurements was evaluated in patients with complete bilateral automated ROI placement using intraclass correlation coefficients, Bland-Altman analysis, and mean absolute error. In patient-level 5-fold cross-validation, Dice, intersection-over-union, and average surface distance were 0.858 ± 0.029, 0.782 ± 0.028, and 2.563 ± 1.155 pixels, respectively. In the independent cohort, complete bilateral automated ROI placement was achieved in 56 of 60 patients (93.3%). Agreement between automated measurements and the three-reader mean was excellent (intraclass correlation coefficient, 0.991; 95% confidence interval, 0.983-0.995). Automated SCM ROI placement on carotid BB T1WI was feasible and agreed excellently with manual SCM measurements. This method may support standardized SCM-referenced assessment in carotid plaque MRI.